bioRxivJClub
An agent-built, verified atlas of public spatial transcriptomics shows that breadth across studies drives transfer and reproducibility
bioRxiv · · Preprint · Open access
Ji, Zou + 8 more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
SpatialDataAgent-DB indexes 2,411 GEO spatial series and verifies 12,163 samples across 12 platforms and 42 species, enabling transfer and reproducibility that improve with study breadth.
- Why it matters: Public spatial transcriptomics deposits vary by lab, and existing catalogues lack a verified, whole-resource denominator, limiting reliable transfer learning and cross-study replication.
- What they did: SpatialDataAgent-DB uses language-model agents to generate conversion code and a frozen rule-based scorer to recompute released values from GEO source files, rejecting all 1,231 corrupted copies in a held-out test.
- The result: In image-to-expression modeling and dementia intent-based search, accuracy and disease-signature replication rise steadily as the number of independent studies increases, with organ explaining 32% of variance versus platform 13% without batch correction.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.
- Open access